CAREER: Development of Adaptive and Efficient Computational Inverse Design Methods for Organic Functional Materials

Texas A&M University, College Station TX

Investigators

Abstract

With support from the Chemical Theory, Models and Computational Methods Program in the Division of Chemistry, Daniel Tabor of Texas A&M University is developing computational simulation and machine learning tools for accelerating the discovery and design of functional materials that interact with light. These materials, which can be built from inexpensive, earth-abundant elements, are needed in future energy storage applications and to build flexible next-generation electronic devices. However, the design of molecules that make up these devices is challenging, as searching for molecules that have all the necessary properties is like searching for a needle in a haystack. To overcome the current challenges in these searches, Daniel Tabor and his research group will integrate machine learning and artificial intelligence methods to build new searching methods that efficiently propose and test new molecules through computer simulations. The Tabor group will develop a set of interactive educational modules to deepen the connection that students have between their understanding fundamentals of light and its role in modern materials science and contemporary issues in data science. The group will develop interactive spectroscopy analysis modules for all levels of instruction, including for high school, undergraduate, and graduate students. Daniel Tabor and his research group will develop a suite of machine learning tools for accelerating the inverse design of organic functional materials, particularly for organic optoelectronic materials and metastable photoacids. The focus of the methods development efforts will be on integrating new physically informed representations for molecular materials with adaptive reinforcement learning algorithms and unsupervised learning methods to form a closed computational discovery loop. The group will build new types of representations for modularly constructed, conjugated materials, implement and test the performance of generative models coupled to reinforcement learning methods on a broad and diverse class of inverse design problems. In addition, the group will expand the utility of unsupervised learning algorithms in chemistry applications, by incorporating chemical information and learning from real-time quantum chemistry characterization of newly identified chemical modules. These simulations predict properties that can be tested in experiments, and the artificial intelligence methods will provide a set of useable rules for what kinds of molecules generally are more useful for electronic applications, empowering chemists to use them in other design applications. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.